Create Embeddings with Confluent Cloud for Apache Flink¶
Confluent Cloud for Apache Flink® supports creating embeddings for Retrieval-Augmented Generation (RAG) workflows from Kafka topics and Flink tables.
An embedding is a numerical representation that captures relationships and meaning within complex data. In a Retrieval-Augmented Generation (RAG) workflow, an embedding helps to retrieve the most relevant information, ensuring AI delivers precise, context-aware results.
In the Confluent Cloud Console, you can choose your preferred embedding model and transform a selected text column in a Kafka topic into a stream of vector embeddings.
In this guide, you use a Flink Action to connect an existing AI model to an input topic that has embedding data and an output topic that contains the embedding.
Prerequisites¶
- Access to Confluent Cloud.
- Access to a Flink compute pool.
- Sufficient permissions to create models. For more information, see RBAC for model inference.
- A Kafka topic to use as a source for the embedding data.
- A model in one of the supported providers. For more information, see Run an AI Model.
Configure the Create Embedding action¶
In the following steps, you select a column in a topic that has embedding data, the name of an output topic that contains the embedding, and the AI model.
Log in to Confluent Cloud Console at https://confluent.cloud/login.
In the navigation menu, click Data portal to open the Data portal page.
In the Environment drop-down, select the environment where you want the embedding to run.
Click the tile of the topic that has your embedding data, and in the details pane, click Actions.
Click Create embeddings.
In the Action details section, select the column to use as the source for your embedding. The dropdown shows all columns in the table that you can use. Only string and VARCHAR columns can be appended to the embedding.
In the Output topic embedding column textbox, enter the name of column in the the Kafka topic that receives the embedding data. The default name is <column-name>_embedding.
In the Model dropdown, select the name of an existing model that was registered previously by using the CREATE MODEL statement, or choose Create new model to register a new model.
To register a new model, provide the following details:
- Model name: Enter the name of the new model.
- Provider: Select one of the supported model providers.
- Endpoint: Enter the URL of the model.
For more information, see Create an AI model.
(Optional) If your data needs to chunked, select Enable chunking and enter values for the related settings.
In the Output topic section, select the Flink catalog and database (environment and cluster) for the embedding table.
In the Table textbox, enter the name of the Flink table that contains the embedding. The default name is <source-topic-name>_embedding.
In the Runtime configuration section, select the Flink compute pool to run the embedding query.
- (Optional) Select Run with a service account for production jobs.
- (Optional) Select Show SQL to view the Flink statement that does the embedding work.
Click Confirm and run to start the embedding statement.